6、OpenMP归约操作reduction、shared、simd

基本思想：OpenMP进行加减乘除运算的规约处理 reduction 可以支持的运算符号 +、-、*、/、&&、|、^、|| 符号

#pragma omp parallel for reduction(+:sum)
    for(int i=0;i<num;i++)
    {
      ....
    }

测试代码

#include <iostream>
#include <omp.h>
#include<chrono>
using namespace std;
using namespace chrono;

void sequentialProgram(int num)
{
    int sum=0;

    for(int i=0;i<num;i++)
    {
        sum=sum+i;
        printf("sum=%d the current thread id: %d\n",sum,omp_get_thread_num());
    }
    std::cout<<"sum="<<sum<<std::endl;

}

void  parallelProgram(int num) {


    int sum=0;

#pragma omp parallel for reduction(+:sum)
    for(int i=0;i<num;i++)
    {
        sum=sum+i;

        printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;


}

int main() {


    int num=omp_get_num_procs();//12

    auto start_time=std::chrono::steady_clock::now();
    sequentialProgram(num);
    auto end_time=std::chrono::steady_clock::now();
    std::cout<<"sequentialProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;

    start_time=std::chrono::steady_clock::now();
    parallelProgram(num);
    end_time=std::chrono::steady_clock::now();
    std::cout<<"parallelProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;
    return 0;
}

测试结果i 可以看出 在增加归约之后，结果和串行程序保持一致，随着叠加次数增加，其时间明显比串行时间的少

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=0 the current thread id: 0
sum=1 the current thread id: 0
sum=3 the current thread id: 0
sum=6 the current thread id: 0
sum=10 the current thread id: 0
sum=15 the current thread id: 0
sum=21 the current thread id: 0
sum=28 the current thread id: 0
sum=36 the current thread id: 0
sum=45 the current thread id: 0
sum=55 the current thread id: 0
sum=66 the current thread id: 0
sum=66
sequentialProgram elapse time: 0.0376821 seconds
i=1 sum=1 the current thread id: 1
i=8 sum=8 the current thread id: 8
i=9 sum=9 the current thread id: 9
i=4 sum=4 the current thread id: 4
i=2 sum=2 the current thread id: 2
i=7 sum=7 the current thread id: 7
i=10 sum=10 the current thread id: 10
i=6 sum=6 the current thread id: 6
i=5 sum=5 the current thread id: 5
i=0 sum=0 the current thread id: 0
i=3 sum=3 the current thread id: 3
i=11 sum=11 the current thread id: 11
 sum=66
parallelProgram elapse time: 0.0163177 seconds

Process finished with exit code 0

同样的逻辑代码

shared 对一个变量的同一个内存进行读写

#pragma omp parallel for shared(sum)
    for(int i=0;i<num;i++)
    {
        .....
    }

simd 编译器可以忽略矢量依赖关系，使循环尽可能实现矢量友好，并尊重用户对同时执行多个循环迭代的意图。

#pragma omp simd
for(int i=0;i<num;i++)
{
......
}

测试代码

#include <iostream>
#include <omp.h>
#include<chrono>
using namespace std;
using namespace chrono;
static omp_lock_t m_lock;
void sequentialProgram(int num)
{
    int sum=0;

    for(int i=0;i<num;i++)
    {
        sum=sum+i;
        // printf("sum=%d the current thread id: %d\n",sum,omp_get_thread_num());
    }
    std::cout<<"sum="<<sum<<std::endl;

}

void  parallelProgram(int num) {


    int sum=0;

    omp_init_lock(&m_lock);
#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
        omp_set_lock(&m_lock);
        {
            sum=sum+i;
        }
        omp_unset_lock(&m_lock);
    }
    std::cout<<" sum="<<sum<<std::endl;
    omp_destroy_lock(&m_lock);


    sum=0;

#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
#pragma omp critical(sum)
        sum=sum+i;

        //printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;


    sum=0;
#pragma omp parallel for
    for(int i=0;i<num;i++)
    {
#pragma omp atomic
        sum=sum+i;

        //printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;


    sum=0;

#pragma omp parallel for shared(sum)
    for(int i=0;i<num;i++)
    {

        sum=sum+i;

        //printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;
    sum=0;

#pragma omp parallel for reduction(+:sum)
    for(int i=0;i<num;i++)
    {

        sum=sum+i;

        //printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;

    sum=0;

#pragma omp  simd
    for(int i=0;i<num;i++)
    {

        sum=sum+i;

        //printf("i=%d sum=%d the current thread id: %d\n",i,sum,omp_get_thread_num());
    }
    std::cout<<" sum="<<sum<<std::endl;

}

int main() {


    int num=omp_get_num_procs();//12

    auto start_time=std::chrono::steady_clock::now();
    sequentialProgram(num);
    auto end_time=std::chrono::steady_clock::now();
    std::cout<<"sequentialProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;

    start_time=std::chrono::steady_clock::now();
    parallelProgram(num);
    end_time=std::chrono::steady_clock::now();
    std::cout<<"parallelProgram elapse time: "<<std::chrono::duration<double>(end_time-start_time).count()<<" seconds"<<std::endl;
    return 0;
}

测试结果

F:\OpenMP\cmake-build-debug\OpenMP.exe
sum=66
sequentialProgram elapse time: 0.0011415 seconds
 sum=66
 sum=66
 sum=66
 sum=66
 sum=66
 sum=66
parallelProgram elapse time: 0.0028554 seconds

Process finished with exit code 0